Comparison of left ventricular strains and torsion derived from feature tracking and DENSE CMR

Figures

• 

  Fig. 1. Representative images with contour overlay from feature tracking (TomTec Imaging Systems) and DENSE in mid-ventricular short-axis (top) and four-chamber (bottom) views. End-diastolic images are shown for both DENSE and feature tracking in a representative subject. In feature tracking (TomTec), only endocardial contours were used for longitudinal strain calculation. Contour-based strains were derived from the same end-diastolic/end-systolic contours exported from TomTec

• 

  Fig. 2. Bland-Altman analyses for circumferential and longitudinal strains between feature tracking and DENSE. Analyses were performed both before (left column) and after (right column) adjusting the feature tracking results to account for differences in the strain calculation. All differences were calculated by subtracting the DENSE strain from the feature tracking strain. All biases and 95% limits of agreement improved after adjusting the feature tracking strains. The red shaded region highlights the bias. The best agreement was observed in mid-ventricular circumferential strain. CoV, coefficient of variation

• 

  Fig. 3. Bland-Altman analyses for torsion and dyssynchrony between feature tracking and DENSE. All differences were calculated by subtracting the DENSE measurement from the feature tracking measurement. The red shaded region highlights the bias. Poor agreement between feature tracking and DENSE was observed for both measures as demonstrated by large biases, 95% limits, and CoVs. CoV, coefficient of variation

• 

  Fig. 4. Bland-Altman analyses for circumferential and longitudinal strains between feature tracking and contour-based strains. All differences were calculated by subtracting the feature tracking strain from the contour-based strain. The red shaded region highlights the bias. Excellent agreement (small biases and tight 95% limits) was observed for all circumferential and longitudinal strains. CoV, coefficient of variation

• 

  Fig. 5. Propagated contours and reported feature tracking strains are inconsistent (representative subject). The propagated endocardial contours for frame 1 and frame 30 (the last frame) of a representative subject are shown along with the strains reported by feature tracking. Despite the differences in contour length, which would be measured as strain by the contour-based calculation, the feature tracking software reported zero strain in all segments and, thus, zero slice-wise strain. When deriving strains, the feature tracking software may employ curve-fitting after propagating the contours, which would lead to differences between the reported feature tracking strains and the contour-based strains